Read the Brief Here

The report comes from the CCC Beyond Code: Engineering Trustworthy Software Systems with AI at Scale workshop held in February 2026, generously supported by the National Science Foundation and the IEEE Computer Society. It brought together experts from academia, industry, government, and national labs to explore how AI-assisted development can produce software that is cost-effective, verifiable, secure, and maintainable. Participants found that current techniques, processes, and training fall short in several areas. These include capturing user intent, meeting quality attributes like security and performance, and keeping review and deployment practices in step with the volume of AI-generated code. Industry participants in the workshop included Amazon, Microsoft, Intel, Atlassian, Peraton Labs, Leibniz Labs, Byggr, and Storer Enterprise Strategy.

Why Industry Action Matters

Industry is at the forefront of adopting AI-assisted development, and is well positioned to shape how it matures. The speed and volume of AI-generated code is outpacing human review capacity, and the bottleneck in software development is moving from implementation to requirements and specification. The brief identifies where industry’s resources and expertise can help address the challenges ahead, from security vulnerabilities to codebases that future teams cannot maintain.

Key Recommendations

The brief offers targeted recommendations for five roles within industry:

• Industry leaders: workforce reskilling, automated quality assurance, intellectual property risk, and legacy code modernization
• Industry partners in academic research: access to frontier compute, the scope of industry-academic partnerships, and foundational research
• Industry researchers: requirement elicitation, formal interfaces for agent communication, and AI-assisted deployment
• Model providers: verification, security, and attribution and training data licensing
• Developer tool and platform builders: IDE design, specification and verification pipelines, traceability, security interfaces, and deployment agents

Read the Brief and Full Report

Read the Key Recommendations for Industry on Trustworthy Software Engineering with AI brief here. For the complete findings and research directions behind each recommendation, we encourage the entire computing research community to read the full Beyond Code report here.

Tune in to the CCC LinkedIn Showcase Page for updates and more resources like this. Stay connected with CCC for the latest insights, publications, and opportunities to engage by subscribing here.

This material is based upon work supported by the U.S. National Science Foundation (NSF) under Award Nos. 2300842 and 2619366. These awards support the Computing Community Consortium (CCC), a programmatic committee of the Computing Research Association (CRA). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Working alongside CCC staff, the workshop was organized by Vandana Janeja (University of Maryland Baltimore County), Manish Parashar (University of Utah), Abhishek Bhattacharjee (Princeton University), Edith Elkind (Northwestern University), David Jensen (University of Massachusetts Amherst), William (Bill) Regli (University of Maryland), and Michela Taufer (University of Tennessee Knoxville). The workshop builds on community input gathered through the AIRE discussion forum and roundtables that took place earlier this year, and was made possible by the support of the National Science Foundation (NSF).

From Challenges to Recommendations

Over 2.5 days, the agenda moved participants from defining problems to drafting solutions. Discussion was organized around the AIRE pillars, the areas of the AI research ecosystem the community identified as under the most strain:

• Publication and peer review
• Infrastructure and resources
• Interdisciplinarity
• Public-private partnerships

The first day began with lightning introductions from every attendee, followed by two rounds of breakouts: one to refine the challenges within each pillar, and one to begin generating solutions. On the second day, an Unconference gave participants room to raise topics of their own before a breakout session focused on solutions and recommendations. The final day turned to prioritizing those recommendations and drafting them pillar by pillar.

Each breakout fed into a full-group report out, so that ideas developed in smaller rooms were tested and built on by the full group.

Perspectives from Across the Ecosystem

Kathy Yelick (Lawrence Berkeley National Laboratory) opened the workshop with a keynote on Science in the Age of SI, setting the stage for the breakout work that followed. And on the second day, Manish Parashar (University of Utah) moderated the panel Framing the Future AI Research Ecosystem with Erwin Gianchandani (NSF), Beth Mynatt (Northeastern University), and Elizabeth Bruce (Microsoft). Together, the keynote and panel grounded the workshop’s discussions in a shared picture of how AI is reshaping research and what the ecosystem will need to keep pace.

Outside of formal sessions, an evening campfire and a scheduled networking hike in the Santa Cruz mountains gave attendees time to continue the conversation informally.

Next Steps for AIRE

The ideas developed in Santa Cruz will be synthesized into a forthcoming CCC workshop report, laying out a concrete set of recommended directions for each pillar of the AI research ecosystem. The report is anticipated to be released in Winter 2026/7. We encourage all members of the computing community to subscribe below to be the first to be notified when it comes out.

Tune in to the CCC LinkedIn Showcase Page for updates and more workshops like this. Stay connected with CCC for the latest insights, publications, and opportunities to engage by subscribing here.

This material is based upon work supported by the U.S. National Science Foundation (NSF) under Award Nos. 2300842 and 2619366. These awards support the Computing Community Consortium (CCC), a programmatic committee of the Computing Research Association (CRA). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

The ruling responds to genuine security concerns. But it also brings into focus how heavily the U.S. robotics research base depends on the very supply chains it is designed to restrict. A critical gap has been brought to light by this ruling: U.S. robotics researchers rely heavily on foreign-made robots and components, and the domestic manufacturing, verification methods, test infrastructure, and workforce needed to replace them are not yet in place.

The Computing Community Consortium’s (CCC) newest white paper, Securing the Robotics Supply Chain: Technical Needs and Investment Opportunities for the Domestic Robotics Research Ecosystem, sets out what a sustained federal research and investment agenda would need to include to close that gap over the coming decade. The paper is authored by Cindy L. Bethel (University of Alabama), Henrik I. Christensen (University of California San Diego), Odest Chadwicke Jenkins (University of Michigan), George Pappas (University of Pennsylvania), and Holly Yanco (University of Massachusetts Amherst).

Securing the Robotics Supply Chain White Paper

What’s Inside

The paper examines where the U.S. robotics research ecosystem stands today, including its reliance on foreign-made research platforms and components, and outlines what it would take to build a secure, domestically capable alternative. Key themes include:

• Verifying trust in a mixed supply chain: Research needed to build and certify robotic systems that can be trusted even when some components come from sources that cannot be fully vetted.
• Investing in hardware and infrastructure: The case for sustained support of component-level hardware research and shared, open-access test facilities.
• Federal investment and coordination: Opportunities to strengthen domestic manufacturing, preserve research access, and move robotics research from lab to market.
• Workforce and education: Building the talent pipeline needed to design, manufacture, and secure robotic systems, from K-12 through the doctorate.

Read the Full White Paper

The response is available now. We encourage all members of the computing community to read the full white paper here. 

Looking for another way to get involved? Drones face many of the same supply chain questions raised in this white paper. On November 3 at 2:00 p.m. ET, CCC will host a Community Chat webinar and Q&A titled Computing on the Fly: Navigating a Vision for the Future of Drone Computing, featuring Dirk Giles (U.S. Forest Service) and Sabine Brunswicker (Purdue University). Learn more and register here.

Tune in to the CCC LinkedIn Showcase Page for updates and more resources like this. Stay connected with CCC for the latest insights, publications, and opportunities to engage by subscribing here.

This material is based upon work supported by the U.S. National Science Foundation (NSF) under Award Nos. 2300842 and 2619366. These awards support the Computing Community Consortium (CCC), a programmatic committee of the Computing Research Association (CRA). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

The Computing Community Consortium (CCC) is thrilled to announce the next webinar in its Community Chat series:

Speakers at this event will include Kevin Butler (University of Florida), Christopher Stewart (The Ohio State University), Nils Aschenbruck (Osnabrück University), Ufuk Topcu (University of Texas), Deborah Silver (Rutgers University), Sabine Brunswicker (Purdue University), and Dirk Giles (U.S. Department of Agriculture Forest Service). The findings are a result of discussions at the CCC Computing on the Fly visioning workshop held December 1-2, 2025 in Washington, D.C. — generously supported by the National Science Foundation (NSF) and IEEE Computer Society — which convened 47 cross-sector experts to articulate a vision for drone development over the next ten years.

Register Here

A New Format for Computing on the Fly

In this webinar, hear from the authors of CCC’s Computing on the Fly: Navigating a Vision for  the Future of Drone Computing workshop report. The authors themselves will present key findings from the report, as well as recommendations tailored to audiences in the research, industry, and policy spheres.

Specifically, hear about topics including but not limited to:

• How to scale to millions of non-military drones
• Standards, certification, and regulation in the drone space
• Workforce development and education
• Advancing AI technology to support a drone infrastructure

These recommendations push us towards a 2035 vision in which drones move goods, medical supplies, and information at a scale comparable to national infrastructure investments like highways and the electric grid. Applications include natural disaster detection drones, medical supply chains that bypass ground congestion to reach rural hospitals, and nationwide fleets that continuously inspect bridges and power lines.

Drone Testing and Wildfire Fighting

CCC is also excited to welcome two phenomenal guest speakers to this webinar and Q&A: 

Sabine Brunswicker, Professor of Digital Innovation at Purdue University and a contributor to the Computing on the Fly report, will present on how we ensure the safety of drones. She founded Purdue’s Center for Artificial Intelligence for Digital, Autonomous, and Augmented Aviation (AIDA3), and in 2024 launched AIrTonomy, a cyber-physical infrastructure program for rapid experimentation and safety verification of AI-enabled autonomy for aviation. 

Dirk Giles, USDA Forest Service UAS Chief, will share how one of the United States’s largest land management agencies puts drones to work. The thermal sensing in drone technology is a well-known use in fire management, but the agency also uses drones for aerial ignition during prescribed and wildland fire operations, particularly in dangerous conditions. Beyond wildfire management, the Forest Service also uses drones across a wide range of missions, including timber, hydrology, reforestation, forest health, and infrastructure inspection.

Join the Conversation

This event is open to the public. We encourage you to share it with anyone who may be interested in the future of drone technology.

Date: November 3, 2026

Time: 2 PM ET

Can’t make it? Register anyway, and we’ll send you a link to the recording afterwards.

We hope to see you there!

Register for Computing on the Fly

Tune in to the CCC LinkedIn Showcase Page for updates and more Community Chats like this. Stay connected with CCC for the latest insights, publications, and opportunities to engage by subscribing here.

This material is based upon work supported by the U.S. National Science Foundation (NSF) under Award Nos. 2300842 and 2619366. These awards support the Computing Community Consortium (CCC), a programmatic committee of the Computing Research Association (CRA). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Watch the Recording Here

Lightning talks from Chandra Krintz (University of California, Santa Barbara), Ana Radovanovic (Google), and Adam Wierman (California Institute of Technology) were followed by a Q&A, joined by Elizabeth Bradley (University of Colorado Boulder), Siddharth Parida (Embry-Riddle Aeronautical University), Masood Parvania (University of Utah), Eve Schooler (University of Oxford), and Bronis de Supinski (Lawrence Livermore National Laboratory).

Two Sides of the Energy Demand Problem

Artificial intelligence (AI) and high performance computing (HPC) are driving advances in healthcare, weather forecasting, and emergency response, but their energy demand has already outpaced what power grids can supply. The International Energy Agency projects that global data center electricity use will more than double between 2024 and 2030. 

At the same time, these technologies play a direct role in addressing environmental problems, supporting work on sustainable agriculture, more efficient batteries, recyclable renewable energy technologies, and sustainable plastics. The authors argue that sustainable computing (reducing computing’s footprint) and computing for sustainability (using computing to address environmental challenges) must be pursued together, producing a multiplier effect that neither achieves alone.

Michela Taufer stressed that efficiency alone is not enough. Individual computations can use less energy even as total consumption rises, because the volume of computing keeps growing. The article calls for research, technology development, and governance to advance together, drawing on computer science, environmental science, power systems, social science, policy, and industry, and closes with open questions for the community.

Three Perspectives on Balancing Sustainability

Which computations are worth doing. Drawing on agriculture and ecology, Chandra Krintz explained that in physical settings, timing matters as much as accuracy: a prediction that arrives after an irrigation or pest management decision has little value. Her team predicts the value of an HPC job before running it. Across roughly 100 jobs on NERSC systems, this improved accuracy gained per core hour by 35% compared with running every job, while using 90% fewer core hours. “The principle is, compute when computation matters,” Krintz said.

When and where to run it. Ana Radovanovic described how data center growth is straining grid capacity and delaying interconnection, making it essential for data centers to be good grid citizens. “Data center fleet growth today cannot happen without a sustainability-aware design throughout its lifecycle, from siting to workload scheduling,” she said, then explaining how Google’s carbon-aware computing shifts flexible workloads in time and location toward cleaner power. Newer approaches include curtailing machine learning training during grid emergencies, partnering with virtual power plant operators, and siting near clean generation.

What it means for host communities. Adam Wierman noted that a majority of Americans now oppose building a data center in their area. “Communities aren’t worried about the long-term trade-offs, they’re worried about what they are experiencing today,” he said. Those costs include noise, rising electricity rates, strain on water systems, and pollution from on-site generators, and they fall disproportionately on low-income communities. Well-designed partnerships can also bring revenue and infrastructure investment, and Wierman sees research as a way to help communities weigh those trade-offs.

From the Broader Discussion

The webinar closed with a Q&A, allowing attendees to ask the questions they cared the most about. This led to a few key discussions related to the broader sustainability question.

What to optimize. Asked what to optimize for if efficiency only drives more consumption, Liz Bradley pointed to the need to assess true cost over the full life cycle, including impacts on disadvantaged communities. That cost is hard to pin down, she noted, because many impacts are easy to overlook and policymakers and investors tend to prioritize the short term, making the right metric “a really hard socio-technical problem.” Radovanovic added that this will take several metrics, including embodied emissions, water use, and how much load can shift when the grid needs it. Krintz also noted that resource data rarely reaches the people deciding what to compute.

Policy and education. Wierman encouraged researchers to show what is possible so policymakers can develop technologically informed policy. Schooler called for teaching metrics beyond performance, from energy and carbon to water and noise, early in computing education.

Standards and disclosure. Asked about report cards for data centers, Schooler pointed to standards work at the Open Compute Project, IEEE, IETF, and ISO. Because standards take time, she suggested pilot projects with community and industry partners as a faster way to put shared metrics into practice. Wierman added that communities need metrics and public disclosure urgently in order to evaluate proposed data centers.

Advice for early-career researchers. Parvania encouraged working on practical problems with industry partners, Krintz urged building interdisciplinary networks early, and Schooler pointed to open questions in the economics of resource use and more decentralized infrastructure.

Watch the Webinar and Read (and Cite!) the Article

If you missed the event, you can watch the recording and read the CACM article below:

Powering the Solution: Balancing Computing’s Energy Demands with Its Contributions to Sustainability Webinar

“Balancing Sustainable Computing and Computing for Sustainability” CACM Article

We hope you will share both resources with colleagues and students working on computing and sustainability, as well as cite them if they prove useful for your own research. For those interested in the topic, we also encourage you to subscribe to CCC’s LinkedIn page or CCC Blog at the links below to be notified about a forthcoming report from the Full-Stack Sustainability workshop in February 2027.

Tune in to the CCC LinkedIn Showcase Page for updates and more Community Chats like this. Stay connected with CCC for the latest insights, publications, and opportunities to engage by subscribing here.

This material is based upon work supported by the U.S. National Science Foundation (NSF) under Award Nos. 2300842 and 2619366. These awards support the Computing Community Consortium (CCC), a programmatic committee of the Computing Research Association (CRA). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Beyond Code: Engineering Trustworthy Software Systems with AI at Scale was authored by Randal Burns (Johns Hopkins University), Sebastian Elbaum (University of Virginia), Gabrielle Allen (University of Wyoming), Nils Aschenbruck (Osnabrück University), Terry Benzel (University of Southern California), William Gropp (University of Illinois Urbana-Champaign), Rick Kazman (University of Hawaii), and Manish Parashar (University of Utah).

Read the Report Here

Avoiding Security Breaches in AI-Generated Software

Software built using AI is entering systems that people depend on. But AI now generates code faster than anyone can review it, and the report finds that the resulting systems carry unclear provenance, configuration errors, unvetted third-party components, and expanded attack surfaces. Faults missed in AI-generated code are already causing security breaches.

Organizations also cannot fully tell what is in the code they receive. Because these systems are black boxes with respect to their training data, users are unable to determine whether generated code contains copyrighted material, infringes patents, or reproduces proprietary work belonging to someone else. Major providers have responded by offering their customers legal indemnification. But interestingly, the market is developing ways to allocate this risk faster than the technical means to actually measure it.

Meanwhile, the checks that would normally catch these problems are weakening:

• Reviewer fatigue. People facing large volumes of machine-generated code miss subtle faults and tend to defer to the machine
• Misleading test results. Automated testing can produce results that look thorough while confirming the wrong behavior
• Concentration risk. When most organizations rely on the same dominant model, they inherit the same blind spots at once

Because of this, verifying code has to scale the way generating code has; and it needs research funding to do so. The report calls for automated checking that runs at machine speed, including formal verification methods that prove a system meets its specification rather than testing samples of its behavior. It also calls for tools that trace generated code back to its sources and for auditing of training data that establishes clear licensing, so organizations can assess what they are deploying before they deploy it. Against concentration risk, it proposes using different AI models to probe each other’s output for vulnerabilities.

Who Can Afford to Study These Systems

Evaluating whether AI systems are safe requires running them at the scale they actually operate. The report finds that the ability to do that now sits almost entirely with industry, which controls the computing power, the training data, and the engineering talent, along with the deployment environments where these systems run in production. Universities increasingly study commercial products after release rather than building the next generation of systems themselves. The report describes academic researchers:

• Evaluating industry outputs rather than developing systems of their own
• Auditing systems they cannot see inside for bias and other properties
• Reproducing findings at a fraction of the scale available to industry labs
• Leaving for industry altogether

University labs still lead on safety questions, including the social and ethical effects of AI, but the report finds academic relevance declining elsewhere. It identifies one fix as fast-acting: giving academic researchers access to frontier computing infrastructure, through a combination of industry funding and national compute initiatives.

A Workforce Built for Different Work

The report finds that the job of a software developer is changing, from writing code toward specifying what a system should do and supervising the AI tools that produce it. Industry has already started hiring on that basis, evaluating engineers on system design and their ability to direct and verify AI output. That shift outpaces the training pipeline. In higher education, the report calls for:

• Curricula centered on requirements, specification, and system design rather than isolated programming exercises
• New teaching approaches, since students will no longer build their understanding of systems through the act of programming
• Retraining beyond the university, reaching the current workforce whose implementation-focused roles are disappearing

Read the Full Report

The findings of this report are informed by discussions at the CCC Beyond Code: Engineering Trustworthy Software Systems with AI at Scale visioning workshop held February 25-26, 2026 in San Francisco, California, generously supported by the National Science Foundation (NSF) and the IEEE Computer Society. The workshop convened 41 experts across academia, industry, and government in artificial intelligence, software engineering, programming languages, cybersecurity, and systems engineering.

Read the Full Report

Tune in to the CCC LinkedIn Showcase Page for updates and more reports like this. Stay connected with CCC for the latest insights, publications, and opportunities to engage by subscribing here.

This material is based upon work supported by the U.S. National Science Foundation (NSF) under Award Nos. 2300842 and 2619366. These awards support the Computing Community Consortium (CCC), a programmatic committee of the Computing Research Association (CRA). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

The workshop will take place October 13-15, 2026, at the Chaminade Resort in Santa Cruz, CA. Over two and a half days, participants will take part in an intensive, laptop-free program focused on consensus-building, working together to define strategic directions and priorities for the AI research ecosystem based on community input.

Understanding the Stressors on AI Research

AI technologies are advancing quickly, and deployed AI systems have become major drivers of economic and social change. Yet the AI research ecosystem itself has also been evolving rapidly.

“In a world increasingly focused on adoption of AI with its myriad uses, potential misuses and concerns around AI security, there is a renewed need to consider all the stressors on the AI Research Ecosystem so that much needed and well thought out AI research can thrive,” says workshop co-chair Vandana Janeja, Professor of Information Systems at the University of Maryland Baltimore County. “I am hopeful that this workshop will lead to recommendations on how best to support this ecosystem into the next several years.”

The workshop’s specific focus areas emerged from months of community input. Through a Consider.it forum, researchers could submit stressors they saw affecting the AI research ecosystem and respond to those raised by others. In parallel, the AIRE Task Force convened virtual roundtables by sector, bringing together perspectives from government (including DoE, NSF, NIH, and NASA), industry (including Microsoft Research), professional societies (including AAAI), and academia (including Carnegie Mellon University and Brown University). The findings from both efforts led to four pillars, which participants will explore in depth at the workshop: 

• Publication and peer review: Conference reviewing systems, journal publishing systems, and pre-publication distribution
• Infrastructure and resources: The computing, data, and funding resources, including Federal funding programs, that make AI research possible
• Interdisciplinarity: How AI research connects with and draws on other fields
• Public-private partnerships: Research translation, translational ecosystems, and standard agreements for industry funding of academic research

“While AI offers amazing opportunities for the academic AI research community, it is also straining the academic research ecosystem, including publication and review structures, academic reward structures, research infrastructure, and partnerships,” says workshop co-chair Manish Parashar, Chief AI Officer at the University of Utah. “I look forward to the AIRE visioning workshop, where we will come together to understand these stressors, brainstorm mitigation strategies, and develop an action plan to restore a vibrant AI research ecosystem.”

Learn More and Stay Up-to-Date

The final sessions of the workshop will focus on synthesizing these discussions into recommendations, ultimately informing a CCC report expected in Spring 2027. 

CCC looks forward to welcoming participants to the workshop. To stay informed about updates and outcomes from the AIRE workshop, follow the CCC Blog and LinkedIn Spotlight Page. We also recommend checking out Janeja and Parashar’s recent Medium post, “Envisioning the Future AI Research Ecosystem” for more information about their initial findings from the field.

This material is based upon work supported by the U.S. National Science Foundation (NSF) under Award No. 2300842 and 2619366. These awards support the Computing Community Consortium (CCC), a programmatic committee of the Computing Research Association (CRA). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

SEMANTiCS is a globally recognised venue where researchers, industry experts, and business leaders exchange knowledge on Knowledge Graphs and Artificial Intelligence. It is a premier space for exploring applications in Neurosymbolic AI, Machine Learning, NLP, and Enterprise Data Management, turning cutting-edge research into practical, real-world solutions.

SEMANTiCS 2026 Blue Sky Winners

First Place

From Fact to Feeling: Narrative Knowledge Graphs for Human-AI Interaction

Anna Sofia Lippolis, Luiz Do Valle Miranda

Second Place

Can a Knowledge Graph Fall Down? Embodied Knowledge Graphs for Humanoid Robots

Blerina Spahiu, Loris Roveda

Third Place

From Sequential Language to Structural Intelligence: Toward Relationship-Aware AI

Alex Romanova

Want to read these papers? They are open access, hosted alongside the SEMANTiCS proceedings. Check them out here!

Please join us in congratulating the SEMANTiCS 2026 winners! 

CCC sponsors many Blue Sky tracks at various conferences in computing every year. If you are interested in requesting sponsorship of a Blue Sky track at your conference, please email blueskyideas@cra.org. 

Requests need only include a brief description of the conference and a proposed list of program committee members for the track. For more information — including guidelines for conference program committees, recommendations for selecting winners, and logistics for issuing CCC-sponsored travel awards to the winners, as well as a sample call for papers for a Blue Sky Ideas track — visit our website.

Tune in to the CCC LinkedIn Showcase Page for updates and more Blue Sky Track announcements. Stay connected with CCC for the latest insights, publications, and opportunities to engage by subscribing here.

Beyond Code: Engineering Trustworthy Software Systems with AI at Scale was authored by Randal Burns (Johns Hopkins University), Sebastian Elbaum (University of Virginia), Gabrielle Allen (University of Wyoming), Nils Aschenbruck (Osnabrück University), Terry Benzel (University of Southern California), William Gropp (University of Illinois Urbana-Champaign), Rick Kazman (University of Hawaii), and Manish Parashar (University of Utah).

Read the Report Here

Critical Paths Forward for Researchers

The report identifies seven technical directions for the computing research community:

• Bridge the intent gap by developing models and workflows capable of interactive requirement elicitation, formalized specifications that capture quality attributes, and traceability mechanisms that record why design choices were made
• Pursue neuro-symbolic integration, pairing generative models with symbolic reasoning so that code and its formal proofs are produced together and verified automatically
• Develop cross-layer orchestration through higher-level languages designed for generative agents, formal interfaces for inter-agent communication, and IDEs that function as orchestration dashboards
• Make security intrinsic to generation through goal-driven security interfaces, secure-by-default models, and cross-AI validation that uses one model to probe another for vulnerabilities
• Establish training data transparency and IP governance, including membership inference, provenance tracking, and models architected for verifiable attribution
• Advance transpilation as a robustness strategy, forcing generated output into safe and verifiable representations and providing equivalence guarantees across hardware platforms
• Extend AI to deployment, developing agents that reason about package contents and dependencies, monitor federated environments, and remediate anomalies without manual intervention

Shifts Specific to Academia

Alongside these technical directions, the report recommends changes to how academic research in this area positions itself:

• Move industry partnerships past benchmark creation
  • Current industrial-academic alliances largely position academia as a supplier of evaluation benchmarks, limiting research impact and constraining the talent pipeline. A deeper partnership would give researchers the production context, telemetry, and scale that building and testing next-generation models and tooling now requires, and give industry a partner working on problems that sit behind corporate walls.
• Aim at the next generation of models rather than evaluating the current one
  • The report suggests building models from the ground up for formal verification, security-by-construction, and long-term maintainability, which is work that corporate labs routinely bypass under market pressure. University labs retain a lead in safety-related work, including the sociotechnical and ethical effects of AI, which gives them an advantage.
• Rebuild curricula around requirements, specifications, and system design
  • Because students will no longer build mental models through the act of programming, new pedagogical approaches are needed that cultivate system-level intuition and complexity management. Computational thinking itself also needs redefinition to include managing higher-level abstractions and anticipating the behavior of agentic systems.

Read the Full Report

The findings of this report are informed by discussions at the CCC Beyond Code: Engineering Trustworthy Software Systems with AI at Scale visioning workshop held February 25-26, 2026 in San Francisco, CA, generously supported by the National Science Foundation (NSF) and IEEE Computer Society. The workshop convened 41 experts across academia, industry, and government in artificial intelligence, software engineering, programming languages, cybersecurity, and systems engineering. 

We encourage all members of the computing research community to read the full report at the link below.

Read Beyond Code: Engineering Trustworthy Software Systems with AI at Scale

Tune in to the CCC LinkedIn Showcase Page for updates and more reports like this. Stay connected with CCC for the latest insights, publications, and opportunities to engage by subscribing here.

This material is based upon work supported by the U.S. National Science Foundation (NSF) under Award Nos. 2300842 and 2619366. These awards support the Computing Community Consortium (CCC), a programmatic committee of the Computing Research Association (CRA). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

The Computing Community Consortium (CCC) is pleased to announce the third webinar in the Community Chat series: Powering the Solution: Balancing Computing’s Energy Demands with Its Contributions to Sustainability. Join the authors of the recent

Guest Speakers Share Their Expertise

In this one-hour Zoom webinar and Q&A moderated by article author Michela Taufer (University of Tennessee Knoxville), hear key findings from the article alongside three short talks centered around different aspects of the sustainability conversation.

• Chandra Krintz, Professor of Computer Science at the University of California, Santa Barbara, will open with a look at agriculture and ecology, and what balancing computing’s demands against broader societal needs looks like in applied environmental research.
• Ana Radovanovic, Senior Staff Research Scientist and Technical Lead for Energy Analytics and Carbon Aware Computing at Google, will follow with carbon-aware scheduling and how industry is thinking through the same balance. She previously led the team behind Google’s Carbon-Intelligent Computing System, which shifts flexible computing workloads to times and places where cleaner power is available.
• Adam Wierman, Professor of Computing and Mathematical Sciences at the California Institute of Technology, will close with a talk on the community impact of data centers, and whether they are a burden or an asset to the communities that host them.

All speakers will then be available for questions from the audience during a Q&A, joined by the other authors of the article:

• Michela Taufer, University of Tennessee Knoxville
• Elizabeth Bradley, University of Colorado Boulder
• Tevfik Kosar, University at Buffalo
• Siddharth Parida, Embry-Riddle Aeronautical University
• Masood Parvania, University of Utah
• Eve Schooler, University of Oxford
• Bronis R. de Supinski, Lawrence Livermore National Laboratory

Join Us

Date: Thursday, October 8, 2026

Time: 3:00 PM ET

The question around computing’s energy usage has become a global one. This is your chance to hear directly from the experts and make your voice heard. Use the link below to register for the webinar, and feel free to submit questions ahead of time. 

Can’t make it? Register anyway and CCC will send you the recording afterward. We hope to see you there!

Register Here

Tune in to the CCC LinkedIn Showcase Page for updates and more Community Chats like this. Stay connected with CCC for the latest insights, publications, and opportunities to engage by subscribing here.

This material is based upon work supported by the U.S. National Science Foundation (NSF) under Award Nos. 2300842 and 2619366. These awards support the Computing Community Consortium (CCC), a programmatic committee of the Computing Research Association (CRA). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.